AI Explainability vocabulary helps readers understand explanation, interpretability, feature attribution, SHAP, LIME and the other concepts behind reliable artificial intelligence. Learning the language of ai explainability makes technical tutorials, model evaluations and practical decisions easier to understand. Instead of memorising disconnected jargon, start by asking what each term describes, what evidence supports it and why it matters.
The core aim of vocabulary mastery for ai explainability is to explain why a model made a particular prediction without mistaking an explanation for proof of causation. A good learner can define a term accurately, compare it with a neighbouring concept, use it in a realistic example and recognise when an apparently confident conclusion goes beyond the evidence.
This article belongs to the eduKateSG Vocabulary Hub and connects to AI Safety Vocabulary.
The core learning map
- Identify: Name the object, measurement or decision involved.
- Distinguish: Separate nearby terms that answer different questions.
- Apply: Use the language in a real scenario.
- Verify: Check evidence, assumptions and limitations.
- Communicate: Explain the conclusion in clear everyday English.
Ten essential ai explainability terms
- explanation — identify where this concept appears in the workflow; then practise using the term accurately in a sentence.
- interpretability — explain what it represents or measures; then practise using the term accurately in a sentence.
- feature attribution — connect it to a real example; then practise using the term accurately in a sentence.
- SHAP — compare it with a related concept; then practise using the term accurately in a sentence.
- LIME — describe its practical consequence; then practise using the term accurately in a sentence.
- saliency — identify where this concept appears in the workflow; then practise using the term accurately in a sentence.
- counterfactual — explain what it represents or measures; then practise using the term accurately in a sentence.
- global explanation — connect it to a real example; then practise using the term accurately in a sentence.
- local explanation — compare it with a related concept; then practise using the term accurately in a sentence.
- faithfulness — describe its practical consequence; then practise using the term accurately in a sentence.
A worked real-world example
A school uses a model to flag pupils who may need extra support. A local explanation highlights attendance and recent assessment patterns; staff must check whether those factors are meaningful, fair and appropriate before acting.
Three distinctions that matter
Interpretability vs explainability
Interpretability concerns how understandable a model is by design; explainability concerns methods used to communicate its behaviour.
Global vs local
A global explanation describes broad model behaviour; a local explanation focuses on one prediction.
Attribution vs causation
A feature contribution to a prediction does not establish that changing that feature will cause the predicted outcome to change.
How to practise
Compare two predictions from the same model. Identify which features the explanation emphasises, then ask whether the explanation is stable and whether a different model could tell a different story.
- Write one sentence defining the main concept.
- Draw a simple process diagram and label its stages.
- Explain the worked example without jargon.
- Identify a limitation or potential source of error.
- Teach the idea to another learner and invite a question.
- Return to the topic after a week using a new example.
Common mistakes to avoid
- Definition without application: knowing a word but not recognising it in a problem.
- Metric without context: reporting a number without its population, baseline or assumptions.
- Correlation without causation: interpreting association as proof of cause.
- One-size-fits-all advice: ignoring the task, dataset and consequences of errors.
Frequently asked questions
What is ai explainability vocabulary?
It is the specialised language used to describe explanation, interpretability, feature attribution, SHAP, LIME, saliency and related decisions.
Which words should beginners learn first?
Begin with explanation, interpretability, feature attribution, SHAP. Learn each in a worked context before adding advanced terminology.
How do I know I understand a term?
You can explain it in your own words, distinguish it from a similar term and use it correctly in a new scenario.
What is the best way to remember the vocabulary?
Use retrieval practice, spaced review and explanation through examples rather than copying definitions repeatedly.
Related vocabulary learning
The AI Explainability vocabulary standard
Mastery means the learner can explain why a model made a particular prediction without mistaking an explanation for proof of causation, and communicate the result clearly enough for someone else to examine the reasoning.